memory trace (Elsey et al., 2018, Kindt et al.,2014; Schiller et al., 2009). Afterwards, 1Hz rTMS
was applied for 15 minutes to a specific brain region, according to the assigned group: l-dlPFC,
r-dlPFC, ctrlOccipital, ctrlSham. For the additional control group (No-reminder), rTMS was
106 The four state-dependent groups expressed comparable levels of SCR during reactivation of
fear memory (ctrlSham: mean SCR to two CS+ presentations ± SD: 0.71 μS ± 0.44 μS; ctrlOccipital: 0.67 μS ± 0.28 μS; r-dlPFC: 0.61 μS ± 0.32 μS and l-dlPFC: 0.75 μS ± 0.49 μS; F3,52 = 0.32, p = 0.81). In addition, fear memory was equally well consolidated in the four
groups, as revealed by the absence of an interaction effect between group and phase (F3,52 =
0.58, p = 0.63). That is, there was no effect of group on SCR that differed between the last four
acquisition trials (day 1) and the two reactivation trials (day 2). These data demonstrate that,
before the reconsolidation manipulation, the conditioned response was equally expressed across
groups.
Day 3. The analysis of SCRs revealed a significant interaction (F8,130 = 2.07, p = 0.043; ηp2 = 0.11) between group (r-dlPFC, l-dlPFC, ctrlSham, ctrlOccipital, and No-reminder), stimulus
(CS+ and CS-) and phase (recall, extinction, and reinstatement).
Specifically, in the memory recall test (48h after fear acquisition), the administration of rTMS
over both right and left dlPFC significantly decreased SCR differences between CS+ and CS-
(r-dlPFC, CS+: 0.60 μS ± 0.40 μS; CS-: 0.51 μS ± 0.38 μS; p = 0.41; l-dlPFC, CS+: 0.65 μS ±
0.30 μS; CS-: 0.54 μS ± 0.32 μS; p = 0.33). Conversely, the expression of fear memory remained stable in the three control groups, with SCRs to CS+ significantly larger than those
to CS- (ctrlSham, CS+: 0.63 μS ± 0.42 μS; CS-: 0.46 μS ± 0.27 μS; p = 0.046; d = 0.74;
ctrlOccipital, CS+: 0.70 μS ± 0.29 μS; CS-: 0.44 μS ± 0.28 μS; p < 0.001; d = 0.99; No- reminder, CS+: 0.80 μS ± 0.25 μS; CS-: 0.67 μS ± 0.28 μS; p = 0.02; d = 0.92).
In the extinction training phase, no significant SCR differences between CS+ and CS- were
observed in any group (ctrlSham, CS+: 0.32 μS ± 0.41 μS; CS-: 0.26 μS ± 0.31 μS; p = 0.77; ctrlOccipital, CS+: 0.36 μS ± 0.23 μS; CS-: 0.23 μS ± 0.16 μS; p = 0.25; No-reminder, CS+: 0.40 μS ± 0.24 μS; CS-: 0.23 μS ± 0.14 μS; p = 0.06; r-dlPFC, CS+: 0.38 μS ± 0.33 μS; CS-:
107 0.33 μS ± 0.24 μS; p = 0.83; l-dlPFC, CS+: 0.34 μS ± 0.24 μS; CS-: 0.27 μS ± 0.12 μS; p = 0.70; see Figure 3). Note that the differential fear response was already eliminated during the
recall phase in the groups that received rTMS over the right and left dlPFC. This result ensures
that the five groups were in a similar state of extinction. Namely, the conditioned fear response
was equally reduced after the extinction training in all groups.
Exposure to the aversive stimulus (i.e., the shock) after fear extinction has been shown to
reinstate the expression of the original fear memory in animals (Bouton, 2002) and humans
(Norrholm et al., 2006). Accordingly, following fear memory reinstatement, we observed
different SCRs between CS+ and CS- in the three control groups (ctrlSham, CS+: 0.62 μS ±
0.46 μS; CS-: 0.31 μS ± 0.30 μS; p = 0.00005; d = 0.73; ctrlOccipital, CS+: 0.52 μS ± 0.24 μS;
CS-: 0.36 μS ± 0.17 μS; p = 0.056; d = 0.67; No-reminder, CS+: 0.61 μS ± 0.32 μS; CS-: 0.32
μS ± 0.21 μS; p = 0.00004; d = 0.86). Crucially, reinstatement was unsuccessful in the right and the left dlPFC groups, in which we observed no difference between CS+ and CS- (r-dlPFC, CS+: 0.59 μS ± 0.41 μS; CS-: 0.48 μS ± 0.29 μS; p = 0.20; l-dlPFC, CS+: 0.46 μS ± 0.24 μS; CS-: 0.44 μS ± 0.31 μS; p = 0.94).
Taken together, these data indicate that rTMS over the dlPFC (after fear memory reactivation)
not only diminished fear expression at recall, but also prevented the return of fear following the
reinstatement procedure. Interestingly, subjective US-expectancy, collected at the end of day 3,
revealed a different pattern. The analysis showed significantly a higher shock-expectancy
(F1,65 = 22.63, p < 0.001, ηp2 = 0.26) for CS+ (mean ratings ± SD: 21.21 ± 23.83) than CS- (7.09 ± 13.06) that was equally present in all groups (F4,65 = 2.09; p = 0.09), thereby indicating no effect of dlPFC stimulation on participants’ learned expectations of the unconditioned stimulus.
108 Figure 3. SCR during memory recall, extinction, and reinstatement phases (day 3). Data are represented as mean ± SEM of the SCR amplitude recorded during memory recall, extinction, and reinstatement phases (Day 3) in the five groups. * denotes significant comparisons (p < 0.05).
109
DISCUSSION
The neural substrates of fear memory reconsolidation in humans remain largely unknown. Here,
to target brain processes implicated in fear memory reconsolidation, we administered rTMS to
the dlPFC – a key area for learning and remembering events (Cabeza and Nyberg, 2000;
Eichenbaum, 2017) – during the reconsolidation time-window of a previously acquired fear
memory.
To ensure the state-dependent efficacy of the treatment, in four groups of healthy
humans, we administered rTMS following a reminder of the fear memory able to trigger the
reconsolidation process (Schiller et al., 2009; Sevenster et al., 2013; Merlo et al., 2014), and
these groups were directly compared with an additional control group in which no reminder
was used (No-reminder). In two of the groups receiving state-dependent rTMS, we targeted the
dlPFC both in the left (l-dlPFC) and right (r-dlPFC) hemispheres; moreover, in the other two
(control) groups, we stimulated the occipital cortex (ctrlOccipital) as an active control site, or
administered sham stimulation (ctrlSham) to control for nonspecific effects of rTMS. Seventy
participants were tested across three days, following established procedures to ensure
acquisition, reconsolidation, extinction, and reinstatement of fear memory (Elsey et al., 2018,
Kindt et al.,2014; Schiller et al., 2009). A physiological measure (i.e., SCR) and subjective
reports (i.e., CS-US contingency ratings) of fear learning and memory were collected
throughout the experiment as dependent measures.
On day 1, results showed that all experimental groups acquired fear conditioning,
demonstrating that fear learning took place equivalently across all participants. On day 2, results
revealed no differences of group on SCR for the CS+ reactivated trials. These data demonstrate
that, before the reconsolidation manipulation, the conditioned response was equally expressed
across groups. On day 3, both l-dlPFC and r- dlPFC groups exhibited decreased physiological
110 both in memory recall and after an extinction-reinstatement phases. In striking contrast, no such
decrease was observed in participants receiving either control rTMS (i.e., stimulation of a
control site and sham stimulation), or dlPFC-rTMS without preceding reactivation of a fear
memory (No-reminder), thus showing both the site-specificity and state-dependency of our
rTMS intervention.
Several alternative explanations of the present findings can be discarded. First, the
results cannot be explained by a general amnestic effect of brain stimulation, as the group
receiving rTMS to the control brain area (occipital cortex) continued to express fear (higher
SCR to CS+ compared to CS-) at recall and following reinstatement. Second, only the
stimulation of the right and left dlPFC was causally associated with no fear response in both
testing phases. Third, the evidence that participants persistently expressed fear (in terms of both
psychophysiological reactions and subjective ratings) when the memory was not reactivated by
presentation of the CS+ confirms that the dlPFC manipulation via rTMS was state-dependent,
and specifically acted on the memory reconsolidation process (Elsey et al., 2018). These results,
together with the absence of fear recovery following reinstatement (Barak and Ben Hamida,
2012), argue in favor of a direct modification of the original memory trace, rather than the
formation of a new memory, as occurs in extinction (Bouton, 2002; Raij et al., 2018).
The present results confirm that the expression of fear, even if successfully
extinguished, can be reinstated by reexposure to the threatening stimulus. Remarkably, we
provide novel causal evidence that this return of fear can be prevented by reactivating the
original memory trace and interfering with dlPFC activity. This highlights the critical role of
the dlPFC in the modification of a previously acquired fear memory. It is widely accepted that
the long-term consolidation of conditioned fear memories depends on plastic changes within
the prefrontal cortex, which exhibits protein synthesis and degradation mechanisms similar to
111 consolidation (Gilmartin et al., 2014). However, the precise role of dorsal prefrontal regions in
fear memories has been generally associated with the conscious appraisal of threat and the
ongoing processing of the fear memory trace (i.e., working memory). Within this debate, the
present results are in line with the idea that the prefrontal cortex plays a key role in the
reconsolidation of memories (Gilmartin et al., 2014; Sandrini et al., 2013; Sandrini et al., 2014;
Sandrini et al., 2015; Kitamura et al., 2017; Javadi and Cheng, 2013; Mungee et al., 2014; James
et al., 2016). When reactivated, memories enter a transient and labile state that can result in the
enhancement or weakening of that specific trace (Agren, 2014). Perturbation of the dlPFC
during the reconsolidation time-window is likely to have altered prefrontal functional
connections not only with the hippocampus – as already postulated for non-emotional memories
(Sandrini et al., 2013; Sandrini et al., 2014) – but also with the amygdala, which is associated
with the fearful component of the reactivated memory trace (Mungee et al., 2014). By
interfering with normal brain activity during the consolidation time-window, the connections
between frontal regions and the amygdala were weakened, thus resulting in decreased fear
expression (Mungee et al., 2014).
Notably, dlPFC stimulation had no effect on the declarative memory about which
conditioned stimulus had been paired with the shock, although this factual knowledge no longer
accounted for reliable fear responses in those subjects. This finding therefore suggests that post-
retrieval stimulation of the prefrontal cortices blocks the reconsolidation of the emotional
component of the memory, while leaving the cognitive component of prior contingency
learning unaffected.
It has to be acknowledged that the idea that brain stimulation can interfere with memory
is not completely new. In 1968, two influential papers reported, in rodents, an elimination of
the fear response by pairing a brief presentation of the conditioned stimulus with an
112 impressive, such an invasive approach could not be easily translated to humans. Crucially our
study identified which neural regions should be the best target for interfering with the memory
consolidation process, which represents a clinical priority. More recent non-invasive
approaches to brain stimulation tried to tackle this issue (Raij et al., 2018; Sandrini et al., 2013;
Sandrini et al., 2014; Javadi and Cheng, 2013; Murgee et al., 2014; Censor et al., 2010; Bernacer
et al., 2013; Javadi and Walsh, 2012). However, none of the aforementioned studies aimed to
reduce fear memories by interfering with the reconsolidation process. Moreover, they failed to
investigate the critical role of the dlPFC in the reconsolidation process, and whether targeting
the right or left dlPFC similarly impacts fear memory – a critical point in the design of clinical
TMS protocols (Karsen et al., 2014). Finally, none of the existing non-invasive brain
stimulation studies tested the strength of the neuromodulation by means of a reinstatement
procedure.
To summarize, these results demonstrate that non-invasive stimulation of the prefrontal
cortex following memory reactivation disrupts the expression of fear to a previously
conditioned threatening stimulus, and argue in favor of a critical role of the dlPFC in the neural
network that mediates the reconsolidation of conditioned fear memories in humans. These
findings provide a step forward toward understanding the mechanisms underlying fear memory
reconsolidation, and they have potential clinical implications for targeting emotional,
maladaptive memories (Pennington and Fanselow, 2018). Uncovering the brain regions
involved in the reconsolidation of emotional memories constitutes a challenging opportunity
for non-invasive brain stimulation and reconsolidation-based interventions, which are
increasingly applied to conditions like phobia, addiction, post-traumatic stress disorder,
114